Quick Takeaways
- Use Starting and Ending Keyframes in Runway when you want a smooth transition between two specific illustrations.
- Keep camera movements consistent across adjacent shots (for example, two slow push-ins rather than alternating rapid pans).
- Always include burned-in or closed captions for viewers watching on mobile with sound muted.
Task Objective & Prerequisites
To create a polished AI video: (1) write a concise script broken into 5-second visual beats, (2) generate 16:9 still keyframes first so lighting and composition are locked, (3) upload those keyframes to Runway and prompt only the camera and subject motion, (4) generate voiceover audio in ElevenLabs, and (5) combine the clips, voiceover, and captions in a timeline editor.
What You Need Before Starting
- •A short 30–60 second script divided into 5-second visual scenes.
- •Still keyframes (generated in Midjourney or captured from approved assets).
- •A Runway account for motion generation and an ElevenLabs account for voiceover narration.
Step-by-Step Implementation Guide
Break Your Script into 5-Second Visual Beats
Draft your spoken narration first (roughly 12–15 words per 5 seconds of speech). Next to each line, write a one-sentence description of the single visual shot that will appear on screen.
Beat 1 (0–5s): Audio: "Most research teams lose hours copying notes between tabs." Visual: Wide shot of a clean desk with a laptop showing three connected research cards.
Generate and Approve Still Keyframes First
Create 16:9 still images for each shot on your list. Inspect and fix any visual issues at the image stage, where iterations are fast and inexpensive.
Keyframe check: Are all 6 shots using the same 16:9 aspect ratio and lighting palette?
Upload Keyframes to Runway Image-to-Video
Open Runway, select Image-to-Video, and upload `shot-01.png` as your starting frame. Instead of re-describing the objects in the picture, write a short prompt describing only the camera movement and subtle motion.
Slow dolly-in camera movement; subtle ambient light shift across the desk; smooth steady motion.
Generate Clean Narration Audio in ElevenLabs
Paste your script into ElevenLabs Studio Projects, select a steady narration voice, and generate the audio track. Regenerate any individual sentence where an acronym or pause needs adjustment.
Script formatting tip: Use an em-dash (—) or period where you want a clean 0.4-second breath pause.
Assemble Clips, Audio, and Captions in a Timeline
Import the 5-second Runway video clips and the ElevenLabs voiceover into your timeline editor. Trim clip transitions to match vocal pauses and add readable captions.
Export specification: 1080p MP4 (H.264), 30fps, normalized -14 LUFS audio.
Common Mistakes & How to Fix Them
Mistake: Subjects morph or distort mid-clip
Fix: Reduce motion intensity and simplify your motion prompt to a single camera direction (such as "slow zoom in") rather than complex multi-action choreography.
Mistake: Running out of video credits before finishing a project
Fix: Never iterate on style inside a video generator; lock style with still images first and generate 5-second previews before upscaling.
Mistake: Voiceover feels rushed against the video cuts
Fix: Keep spoken scripts under 130 words per minute and insert 0.5-second room-tone pauses between scene transitions.
Frequently Asked Questions
Why not generate video directly from text without images?
Text-to-video has to guess both the visual design and the motion simultaneously, which leads to inconsistent style across shots and higher credit waste.
How long are individual clips from AI video tools?
Most generative video models generate 4 to 10 seconds per clip, which you then sequence in a video timeline.
Can I dub my finished video into another language?
Yes. Once your English video is exported, you can upload it to ElevenLabs Dubbing Studio to translate and re-voice the audio track.